Luca Schmidtke
Luca Schmidtke
Hi! The checkpoint and inference code are released. You can find it in `pretrained_model.py`.
Hi! The code should produce the same results of the provided checkpoint. How do the reconstructed images look like? The model was trained on roughly 600k images.
Okay, the reconstructions don't look good. What's your batch size and learning rate? Also, how many epochs are you training for?
That seems reasonable to me. I will run the code myself with those parameters and I'll get back to you.
Hi, I encountered the same issue, I suppose it's due to a newer version of one of the packages. I will try and find the cause.
Hi, sorry for the late reply. The `Predictor` class in `pretrained_model.py` returns tensors containing the xy-coordinates for the anchorpoints of all transformed parts. There are duplicates in there because some...
You can interpret the dimensions like this: [batch, number_of_parts, anchor_number, xy(z), ()], where () are dimensions added for broadcasting matrix mutliplications. So the first tensor with shape [1,,1, 3, 3,...
Hi, I will update the repo, just need a little more time.
Hi, it's 600k training images in total, but with random frame pairs.
Okay, so I counted the number of possible pairs in each bounding box bucket for each subject and each activity. Number of possible pairs would be (num_frames choose 2). I...